An Investigation of the Costs and Benefits of Thinning on the Straight Line Hough Transform

Abstract

This paper presents the outcome of an investigation of the costs and benefits of thinning as part of preprocessing for line detection including specification of end-points, from visual images of indoor rectilinear environments. This is done as part of a bigger process with the goal of detecting lines to enable a small mobile robot self-navigate within the environment based on navigationally important features such as doors and corridors reconstructed from the lines detected. The straight line Hough transform is used to determine parameters which specify gradients and positions of lines, and then the end-points of the lines are determined. To do this images can be preprocessed to the point of edge-detection which typically yields edge lines several pixels thick, or edge-detection followed by thinning yielding edge lines about a single pixel in thickness. Since the Hough transform operates on all pixels in an input image, more work is needed to process the "unthinned" image. However, thinning itself takes time. This paper looks into whether the taking the time to do thinning is justified in terms of overall time taken, and quality of resulting lines found, and concludes that for the purpose described, thinning does appear to improve the quality of line detection, while taking less total time to do it.

Corridors and Doors Recognition from Detected Lines and Vanishing Point

“…Figure 2 shows the same image after edges have been detected using the Sobel edge detection operators described in [5] and thinned using a method introduced in [1]. [6] has established that for the vision system developed in this work, thinning yields higher quality line detection while ultimately saving processing time, despite the time taken to do thinning itself. [2] take a thinned binary image such as the one in figure 2 and establish parameters indicating the angle of each important line in the image to the vertical, and its distance to the centre of the image which is taken as the origin, i.e., the point (0,0).…”